The group-III nitrides, for example GaN, are promising wide band-gap semiconductors for optical devices in the blue and ultraviolet (S. Nakamura, 1997a, 1998b), high-temperature and high-power device applications (H. Morkos et al., 1994; T. P. Chow et al., 1994, J. C. Zolper et al., 1996).
However, the reliability of current state-of the art GaN-based devices such as blue emitter and high-temperature devices is limited. Blue diode laser lifetimes are short. This is widely attributed to the fact that most GaN devices are grown on lattice mismatched substrates such as sapphire which results in high dislocation densities, typically about 10.sup.10 cm.sup.-2.
Epitaxial GaN films have recently attracted much interest based on their optoelectronic applications as blue-ultraviolet optoelectronic devices and high temperature transistors (S. Nakamura, 1997, 1998). Since bulk GaN substrates are not available currently, the films are generally grown on sapphire, SiC, GaAs, or Si substrates. These substrates provides poor lattice and thermal expansion matching to GaN which lead to very high densities of structural defects. The identification of an appropriate matched substrate for epitaxial growth might enable the preparation of high quality devices with these semiconductor materials.
One growth process for preparing single crystal films of GaN relies on the vapor phase reaction between GaCl.sub.3 and NH.sub.3 in a hot-walled reactor, such as a halide vapor phase epitaxy (HVPE) system. Sapphire substrates are often used because they are readily available. However, since sapphire is not lattice matched to GaN, GaN and sapphire have very different thermal expansion coefficients. Accordingly, the resulting GaN has poor crystalline quality, having high dislocation densities and other lattice imperfections. Even so, growth of GaN on sapphire is still common. Furthermore, attempts have been made to reduce the occurrence of these dislocations and other lattice imperfections by providing buffer layers, such as AlN or ZnO, between the sapphire and the GaN. However, the defects from the substrate mismatch propagate through the buffer layers to the GaN film.
Though the first demonstration of the fabrication of single crystal GaN occurred 30 years ago, interest in these materials for real-world optoelectronic devices has grown only in the last 5-6 years as material quality has improved and controllable p-type doping has finally been achieved.
A primary difficulty in producing high quality GaN single crystal has been the lack of lattice matching substrates such that high quality GaN single crystal epitaxial films could not be produced. Since high quality bulk GaN substrates have not been available, GaN films are generally grown on sapphire, SiC, or Si substrates. However, III-V nitride compounds having the wurtzite structure which is hexagonal in symmetry, in general, have much smaller lattice constants (a-axis dimension=3.104 .ANG. for AlN, 3.180 .ANG. for GaN and 3.533 for InN) as compared to the currently available semiconductor substrates which typically have cubic symmetry. Accordingly, sapphire, SiC, and Si provide poor lattice, as well as thermal, matching to GaN which can lead to very high densities of structural defects.
The first blue GaN-based light emitting diodes (LEDs) and lasers, are now commercially available. They are fabricated from epitaxial GaN grown on sapphire substrates (S. Nakamura, 1997). The best published lifetime for a GaN-based laser on sapphire is only tens of hours, probably due to the high density of crystal defects. Recently a laser lifetime of 10,000 hours has been reported for devices fabricated on lateral overgrowth epitaxial material on patterned sapphire substrate (S. Nakamura, 1997a, 1998b, 1998c).
Apparently the GaN that laterally overgrows on the SiO.sub.2 mask (in between the mask openings) has dislocation densities that are orders of magnitude lower than material grown directly on sapphire. Lasers fabricated in this low defect density material have much longer life time.
This epitaxially laterally overgrowth (ELOG) technique involves the growth of a GaN buffer layer on a substrate of, for example, Si, GaAs, Sapphire or SiC. A pattern of SiO.sub.2, for example, stripes, is then grown on the GaN buffer layer. The SiO.sub.2 are about 0.2 .mu.m in thickness and preferably covers about two-thirds of the buffer layer. An example may have 6-8 .mu.m wide SiO.sub.2 stripes with 4 .mu.m spacing. As the growth of GaN is continued, the GaN does not grow on the SiO.sub.2 stripes but, rather, only in the grooves. As the GaN growth in the grooves reaches the height of the SiO.sub.2 stripes, the GaN continues to grow up, but also begins to grow laterally from the sides of the GaN ridges to eventually form one continuous film. The defect density of the ELOG GaN film can be on the order of 10.sup.7 /cm.sup.3, with a reduced number of threading dislocations in GaN layer compared with GaN grown directly on the substrate without the SiO.sub.2 stripes. The original substrate material can then be removed, for example via etching, but the SiO.sub.2 grooves are still trapped inside the GaN material. Furthermore, removal of the original substrate can damage the GaN material.
Recently in GaN multi-quantum-well-structure laser diodes (LDs) grown on GaN substrates were demonstrated (S. Nakamura, et al., 1998). The LDs showed a lifetime longer than 780 h despite a large threshold current density. In contrast, the LDs grown on a sapphire substrate exhibited a high thermal resistance and a short lifetime of 200 h under room-temperature continuous-wave operation.
Because of high dissociation pressure of nitrogen over GaN (&gt;70 kbar at 2300.degree. C.), no one has succeeded in making large bulk GaN single crystal substrates. Currently, bulk crystals with dimensions of only a few millimeters can be obtained with high pressure synthesis (20 kbar and 1600.degree. C.) (I. Grzegory et al., 1993) and by hydride vapor phase epitaxy (HVPE) on SiC or sapphire substrates with subsequent substrate removal by reactive ion etching, laser pulses, or by polishing. Accordingly, GaN is usually made by heteroepitaxy onto lattice mismatched substrates such as sapphire (S. Nakamura, 1997a) and silicon carbide (Yu V. Melnik et al., 1997) with subsequent substrate removal by reactive ion etching or wet chemical etching, by laser pulses or by polishing. Each of these removal procedures can cause residual strain, changes in chemical composition of epitaxial films, etc. In addition, not only are the lattice constant of the GaN film and substrate very different, but so are the thermal expansion coefficients, creating additional inducement for the creation of dislocations. These two factors, lattice constant mismatch and very different thermal expansion coefficients, can result in GaN epitaxial films with high densities of dislocations (10.sup.10 I cm.sup.-2, regions of built-in strain, and cracks which often occur due to thermal stress during cooling.
This suggests that if low dislocation density bulk GaN substrates were available, device life times approaching the 50,000 target for reliable CD-ROM storage devices could readily be achieved. Similar improvements can be expected with respect to the reliability of other GaN-based devices such as heterojunction bipolar transistors and modulation-doped field effect transistors for high-temperature electronics and uncooled avionics.